osec.one case analysis cover: Hot Topic and Snowflake 2024, a vendor's infostealer log, 350 million claimed, 57 million counted

In October 2024 a forum seller offered data on 350 million Hot Topic, Box Lunch and Torrid customers. Three weeks later Have I Been Pwned loaded 56,904,909 unique email addresses from it, about 16% of the claim. The security firm Hudson Rock had meanwhile traced a likely way in: password-stealing malware on a computer belonging to an employee of Robling, a retail analytics company, holding credentials for Hot Topic's and Torrid's Snowflake and Looker environments.

That story is usually told as settled. It isn't. We could not find an account of it published by Hot Topic, and the part everyone repeats, a Snowflake account without MFA, comes from the seller. This write-up separates what was verified from what was claimed, works through the numbers, and gives the checks you can run on your own warehouse, including who else's laptop can log in to it. For the wider campaign that year, see our analysis of the 165 Snowflake customers.

Six steps as reported: stealer on a Robling employee's computer on 12 September 2024; over 240 credentials, dozens tied to Hot Topic and Torrid on Snowflake and Looker; a login without MFA, claimed by the attacker and never confirmed; a 730 GB database dated about 19 October; forum sale from 21 October claiming 350 million; 56,904,909 accounts verified by Have I Been Pwned
The reported path. Steps 1 and 2 are Hudson Rock's findings, step 3 is the attacker's claim, and steps 4 to 6 come from the listing and third-party analysis of the data.

Who has established what

StatementWho says itStanding
Data on 56,904,909 accounts is realHave I Been Pwned: marked verified, breach date 19 Oct 2024, added 11 Nov 2024Strongest fact here. Verified as a data set, with no disclosure link
The database is 730 GB and runs from 2011 to October 2024Atlas Privacy, via BleepingComputer and CyberInsiderAnalysis of a copy. Atlas was not 100% certain it is Hot Topic's, though about half the addresses were new ("nearly half" in BleepingComputer, "over 55%" in CyberInsider)
An employee at Robling was infected on 12 Sep 2024Hudson Rock, from its Cavalier stealer-log dataEvidence published as screenshots. Hudson Rock itself says it doesn't conclusively prove how the breach happened
The same username was used in the attackHudson Rock: the seller gave a username matching one on that computerStrong link, but the seller is the only source for how it was used
No MFA on a Snowflake account, plus "other links"The seller, relayed by Hudson Rock ("we emphasize, the hacker CLAIMED")Unverified. Hudson Rock calls it likely true on the surrounding evidence
350 million customersThe sellerNot supported by any count of the data we found
Hot Topic's own accountHot TopicNone found. No response to Hudson Rock, BleepingComputer, TechCrunch or SecurityWeek at the time

We also checked the regulators' side on 4 November 2026. California's breach list holds two Hot Topic filings, both for the 2023 login attacks covered below, and no row for October 2024 under Hot Topic, Torrid, BoxLunch, Box Lunch or Robling. Have I Been Pwned's own record still shows no disclosure link. That isn't proof that no notice exists anywhere, since we didn't check every state, but it is what the public trail shows.

Also note what doesn't connect Robling to fault. Nobody has shown that Robling's own systems were breached; the finding is about one person's computer. Hudson Rock's report says it doesn't imply responsibility for any party. Reports vary on the person's role too: "employee" of Robling in Hudson Rock and PCMag, "a developer with access to Snowflake" in 404 Media's account of what a hacker said, and "a Hot Topic employee" in SecurityWeek, which doesn't match the original report.

Timeline, counted from the infection

DateEventDaySource
12 Sep 2024Infostealer infection on the Robling employee's computer0Hudson Rock
19 Oct 2024Apparent date of the breach37Atlas Privacy, HIBP
20 Oct 2024A hacker calling themselves Dark X tells 404 Media of stealing data on 350 million customers; the next day the data is listed, and the day after that the hacker says Hot Topic kicked them out38404 Media (attacker's word; we read only the free part of the article)
21 Oct 2024"Satanic" lists the data on BreachForums: $20,000, or $100,000 from Hot Topic to take the thread down39Hudson Rock, BleepingComputer
23 Oct 2024Hudson Rock publishes the infection link41Hudson Rock
By 7 Nov 2024Atlas Privacy has a copy and opens a free lookup56Hudson Rock's update note
11 Nov 2024HIBP loads 56,904,909 accounts; asking price now $4,00060HIBP, BleepingComputer
13 Nov 2024Asking price $3,500; Hot Topic has not notified customers or state attorneys general62TechCrunch
Oct-Dec 2024Class actions filed, two also naming SnowflakeTruth in Advertising's case list

If this infection was the source, the 37 days are the part defenders can act on. That would be a fresh theft, not a stale password: in the wider Snowflake campaign Mandiant found infections dating back to November 2020. And 19 October is 131 days after Mandiant's report of 10 June 2024 told every Snowflake customer about this pattern.

Would routine rotation have closed it? A 90-day rotation has a 37 / 90 = 41% chance of landing inside a 37-day window, assuming a random start. A monthly one always would. But that only helps if the malware is gone, because a stealer still running simply takes the new password. Nothing published says whether this infection was a single grab or kept running, so rotation is a backstop at best.

From 350 million to 57 million

The claim moved before anyone counted. Hudson Rock's CTO told PCMag that the seller first boasted of 1 billion users. "I asked him how many uniques, he said 350m and later edited the thread as well."

CountFigureShare of the 350 million claimSource
First boast1 billion usersAlon Gal to PCMag
Forum claim350,000,000 customers100%Hudson Rock, on the thread
Rows indexed in the data384,099,181110%DataBreach.com (Atlas Privacy's lookup site)
Unique email addresses, HIBP56,904,90916.3%HIBP
Email addresses, Atlasabout 54.5 million15.6%DataBreach.com; SecurityWeek says about 54 million

The claim overstates HIBP's count 6.15 times (350,000,000 / 56,904,909). Our reading, not anyone's finding: 350 million looks like a count of rows, not people. Rows in the data come to 384 million, or 6.75 for every account, which is what you'd expect once purchases, payment methods and points each take their own table. It would also explain how "billions of payment details" in the thread sits beside about 25 million card entries in the analysis. An email address isn't quite a person either, so even 57 million is a count of accounts.

Atlas's breakdown shows how thin many of those records are. Not every field exists for every person:

FieldRecordsPer 100 email addresses
Email addressesabout 54.5 million100
Credit card entriesabout 24.9 million46
Namesabout 24.8 million46
Phone numbersabout 24.7 million45
Home addressesabout 9.9 million18
Job titlesabout 344,0000.6

One caution on that source: the DataBreach.com page, as we read it, gives an estimated attack date of 18 November 2023 and describes the 2023 login attacks, which doesn't fit an October 2024 data set. We use its field counts because SecurityWeek reports the same order of magnitude from Atlas, and we use HIBP for the date. The size has two versions as well: 680 GB, 116 GB of it customer data from the seller, and 730 GB from Atlas. At 730 GB and 56.9 million accounts that is about 12.8 KB an account, so the bulk is plainly not contact details. CyberInsider's first report mentions tax records, invoices and shipping details among what was claimed.

What the infected computer held, and what nobody has published

Published by Hudson RockNot published anywhere we could find
Over 240 credentials on the machine, many corporateWhich stealer family it was
Dozens of credentials on corporate URLs for Hot Topic and Torrid on Snowflake and LookerWhether the computer was a personal or a company machine
Browsing history pointing at Hot Topic and Torrid work on snowflakecomputing.com, azure.com and looker.comHow the infection arrived
A username primarily linked to Snowflake URLs for both retailers, matching the one the seller gaveWhether session cookies were taken or used
Many Hot Topic and Torrid email domains among the seller's samplesWhich account was used, what role it held, and how the data left

Two popular retellings go past this evidence. One is that a contractor's personal laptop did it. It may have, and Mandiant did see contractor laptops used for gaming and pirated software in the wider campaign, but no source says so here. The other is that session cookies beat MFA. There's no published sign of that either: the claim is the simpler one, that no MFA was there. If a second factor did exist, stolen session cookies can get around it, which is why the controls below don't stop at MFA.

It's worth asking why one person's browser could hold that much. A vendor that works inside clients' warehouses commonly gets a login in each client's environment. Each client treats that as one more user, while the vendor's laptop ends up as a keyring across clients. Here it held credentials for both Hot Topic's and Torrid's environments.

What was in the card data

Sources describe the same field differently, and the differences matter if you're assessing your own exposure:

SourceWhat it says about cards
Seller, via Hudson RockLast four digits, card type, hashed expiry date, account holder name
HIBPPartial card data: card type, expiry, last four digits
Atlas, via BleepingComputer25 million card numbers encrypted with a weak cipher that's easy to break
Atlas, via CyberInsiderAbout 25 million records with partial details, partly encrypted with outdated protocols

None of these describes verified full card numbers in the clear. The hash is the instructive part, because an expiry date barely has any values to hide. With 12 months and, say, 20 years that's 240 candidates, so hashing it protects almost nothing. If each row had its own random salt, the whole set would still take only 240 x 24.9 million = about 6 billion hashes, or around 100 minutes at an assumed 1 million hashes a second. How these values were actually hashed hasn't been published, but the sum holds for any fast hash. The same goes for the last four digits, with 10,000 values. For fields this small, only a keyed function or encryption with the key held outside the warehouse helps, and even that protects nothing against an account allowed to read the decrypted view.

Two earlier warnings on the same accounts

The seller's listing also included points: "billions of loyalty points" linked to profile IDs, many with no expiry, according to Hudson Rock, which warned of account takeovers. HIBP's field list doesn't include points. The risk wasn't theoretical at this company, even though the route was different:

Attack dates (2023)DaysReported to CaliforniaDays after the last attack
7 Feb, 11 Mar, 18-21 May, 27-28 May, 18-21 Jun1231 Jul 202340
18-19 Nov, 25 Nov328 Mar 2024124

Both notices describe automated logins on the website and app with valid credentials from "an unknown third-party source", on 15 days in all. Hot Topic said it wasn't the source of those credentials and that it couldn't tell attackers' logins from customers' own. It added bot protection after the first set, and the second came anyway. The first notice's text lists 19-21 May while California's list has 18-21 May, so the notice itself adds up to 11 days and the list, like the press reports, to 12. Neither notice gives a number of people.

A dump of loyalty profiles is a ready-made target list for that kind of attack. The defence is the one in our Chick-fil-A analysis: check at the risky moments, such as an email change or a first redemption, not just at login. For the sign-in side, Login Protection adds abuse rules without a large contract.

What the data sold for

Price pointAmountPer 1,000 accounts (56,904,909)
Ransom to remove the thread$100,000$1.76
First asking price, 21 Oct$20,000$0.35
Asking price, 11 Nov$4,000$0.07
Asking price, 13 Nov$3,500$0.06

That's an 82.5% fall in 23 days. These are asking prices only: nothing shows that anyone paid. The direction is what you'd expect once a data set has started to circulate. It also means the data's cost is no barrier to anyone planning a phishing run against these customers.

How it differs from the cases around it

Hot Topic 2024Snowflake campaign 2024Disney 2024
Whose deviceA vendor's employee, per Hudson RockSeveral contractor laptops, some also used for gaming and pirated software, per MandiantAn engineer's home PC
Age of the theft37 days before the breach, if this infection was the sourceInfections back to Nov 2020; at least 79.7% of accounts previously exposedStolen through a trojanised add-on on that PC
TargetWarehouse and BI environments of two retailersAbout 165 organisations notified1.1 TB of internal Slack
Public recordNo company statement found; lawsuits, unprovenMandiant's investigationGuilty plea, per the US Attorney

Run these on your own Snowflake

These use columns from Snowflake's ACCOUNT_USAGE views. We checked the names against the documentation but haven't run them on a live account, so try them first in a non-production role. LOGIN_HISTORY keeps 365 days and can lag by up to 2 hours.

Who can still log in with only a password?

sql
SELECT name, email, type, has_password, has_mfa,
       password_last_set_time, last_success_login
FROM snowflake.account_usage.users
WHERE deleted_on IS NULL
  AND has_password = TRUE
  AND has_mfa = FALSE
ORDER BY password_last_set_time;

Who actually did it in the last 30 days, and from where?

sql
SELECT user_name, client_ip, reported_client_type,
       COUNT(*) AS logins, MAX(event_timestamp) AS last_seen
FROM snowflake.account_usage.login_history
WHERE is_success = 'YES'
  AND first_authentication_factor = 'PASSWORD'
  AND second_authentication_factor IS NULL
  AND event_timestamp > DATEADD(day, -30, CURRENT_TIMESTAMP())
GROUP BY 1, 2, 3
ORDER BY logins DESC;

Pin a vendor's user to its network and to SSO

The address range is a documentation example, and the SSO policy assumes you already have a SAML integration. Test on one user first: Snowflake warns that restricting methods can break drivers and third-party integrations.

sql
CREATE NETWORK RULE vendor_a_rule
  MODE = INGRESS TYPE = IPV4
  VALUE_LIST = ('203.0.113.0/24');
CREATE NETWORK POLICY vendor_a_policy
  ALLOWED_NETWORK_RULE_LIST = ('vendor_a_rule');
ALTER USER vendor_a_analyst SET NETWORK_POLICY = vendor_a_policy;

CREATE AUTHENTICATION POLICY vendor_a_auth
  AUTHENTICATION_METHODS = ('SAML');
ALTER USER vendor_a_analyst SET AUTHENTICATION POLICY vendor_a_auth;

A network policy on a user overrides the one on the account, so a loose user policy can quietly undo a strict account policy. Check both.

Snowflake's own deadline doesn't finish the job

Snowflake's rollout now reaches the gap claimed here. Its documentation gives these estimated dates, and each account is told its own within the rolling windows:

PhaseEstimated datesWhat changes
1Sep 2025 to Jan 2026Password sign-ins in Snowsight need a second factor. BI tools and similar aren't covered
2May to Jul 2026New human users need a second factor; no new LEGACY_SERVICE users
3Aug to Oct 2026All human users need a second factor; non-human users can't use passwords

So by now a password alone should no longer open a normal Snowflake account. That leaves three gaps. Reader accounts, trial accounts and Snowflake Postgres are excluded, and a reader account is something you might well set up for a vendor. A stolen session or a stolen key pair isn't stopped by a password rule. And the same vendor's Looker, Azure and every other login on that keyring sit outside Snowflake's rollout altogether.

Controls that break each step

The starting points are our suggestions for a mid-size company, not published standards.

StepControlStarting point
Stealer on a vendor's computerContract terms: managed devices only for client data, endpoint protection, notice of infectionNotice within 24 hours. Or give access only through a virtual desktop you control, so credentials aren't in the vendor's browser
Credentials in a browser storeSSO from your identity provider or the vendor's, no saved passwords, passkeys where offeredNo password-based human user for any vendor
Late discoverySearch stealer-log data for your own domains, your tenant URLs and your vendors' staffCheck weekly at least: a 37-day window gives you several chances. On a hit, revoke sessions and clean the machine before any reset
Login from a new placeNetwork policy per vendor user, alert on a new client IPAlert on the first login from any new address, no threshold
Too much read accessVendor roles limited to the tables the work needs, masked customer fields, no card fields, secure data sharing in place of a loginReview each vendor role every quarter and after each contract change
Bulk exportAlert on large unloads and `GET` calls by user. That's how the wider campaign took data; how this data left hasn't been publishedAlert on any export by a vendor user over 1 GB. At 730 GB, a week of steady exfiltration is 104 GB a day
Old data held foreverRetention limits. This set reportedly ran back to 2011Delete or archive customer rows with no activity for 3 years. Thirteen years of history is 13 years of exposure
Silence afterwardsDecide now who answers a forum claim, and a template for customersFirst public statement within 72 hours of a credible sample

The retention row may be the most underrated. MFA decides who gets in; retention decides how much is there when someone does.

If you were a customer

• Check your address on Have I Been Pwned.
• Expect phishing that quotes your real orders, card type and last four digits. Those details don't prove a message is genuine.
• Change the password on your rewards account and anywhere you reused it, given the 2023 login attacks.
• Check your points balance. Points show up on no bank statement, so no one else will notice them leaving.

Limits of this write-up

• The path is Hudson Rock's reconstruction plus an attacker's statements. It may be incomplete or wrong in part, and the seller's "other links" are unexplained.
• We found no statement from Hot Topic or Robling. If one exists, it would outrank this reconstruction.
• The court cases are allegations, unproven. Truth in Advertising lists nine suits, eight of them shown as pending when we read it on 4 November 2026 and one voluntarily dismissed. We did not read the dockets, so current status may differ.
• We don't know how much data came from Hot Topic itself and how much from Torrid or Box Lunch.

This won't protect a data warehouse by itself, but it does take one password off the list for your own app: email codes, Google and Apple sign-in, with nothing for a stealer to lift from a browser's password store and no OAuth app to set up. Free.

Add passwordless sign-in

Sources

• Hudson Rock (Alon Gal), Largest Retail Breach in History (23 Oct 2024, with a later update): https://www.infostealers.com/article/largest-retail-breach-in-history-350-million-hot-topic-customers-personal-and-payment-data-exposed-as-a-result-of-infostealer-infection/
• Have I Been Pwned, Hot Topic record (read 4 Nov 2026): https://haveibeenpwned.com/Breach/HotTopic
• BleepingComputer, HIBP notifies 57 million people of Hot Topic data breach (11 Nov 2024): https://www.bleepingcomputer.com/news/security/hibp-notifies-57-million-people-of-hot-topic-data-breach/
• TechCrunch, Hot Topic data breach exposed personal data of 57 million customers (13 Nov 2024): https://techcrunch.com/2024/11/13/hot-topic-data-breach-exposed-personal-data-of-57-million-customers
• PCMag, Hacker May Have Breached Hot Topic, Stolen Data on Millions (updated 24 Oct 2024): https://au.pcmag.com/security/107921/hacker-may-have-breached-hot-topic-stolen-data-on-millions
• 404 Media, Inside the massive crime industry that is hacking billion dollar companies (4 Nov 2024; free portion only): https://www.404media.co/inside-the-massive-crime-industry-that-is-hacking-billion-dollar-companies-copy-2/
• SecurityWeek, Millions of Hot Topic Customers Impacted by Data Breach (12 Nov 2024): https://www.securityweek.com/millions-of-hot-topic-customers-impacted-by-data-breach/
• CyberInsider, first report (23 Oct 2024): https://cyberinsider.com/hot-topic-allegedly-breached-350-million-customers-data-for-sale/ and verification (11 Nov 2024): https://cyberinsider.com/hot-topic-data-breach-impacting-57-million-customers-verified/
• DataBreach.com (Atlas Privacy), Hot Topic: https://databreach.com/breach/hot-topic
• RH-ISAC summary (24 Oct 2024): https://rhisac.org/threat-intelligence/infostealer-infection-results-in-one-of-the-largest-retail-breach-in-history/
• Hot Topic customer notices at the California AG, November 2023 attacks: https://oag.ca.gov/system/files/Hot%20Topic%20Notice%20of%20Data%20Breach.pdf and February to June 2023 attacks: https://oag.ca.gov/system/files/Hot%20Topic%20Inc.%20-%20Notice%20to%20Consumers.pdf
• California AG breach list, searched by name 4 Nov 2026: https://oag.ca.gov/privacy/databreach/list
• Cybersecurity Dive, 12 days of breaches (Aug 2023): https://www.cybersecuritydive.com/news/hot-topic-12-breaches-credential-stuffing/689718
• Security Affairs, November 2023 attacks (29 Mar 2024): https://securityaffairs.com/161192/data-breach/hot-topic-suffered-credential-stuffing-attacks.html
• Truth in Advertising, Hot Topic class actions: https://truthinadvertising.org/class-action/hot-topic/?pg=5
• Mandiant, UNC5537 targets Snowflake customer instances (10 Jun 2024): https://cloud.google.com/blog/topics/threat-intelligence/unc5537-snowflake-data-theft-extortion
• Snowflake documentation: MFA rollout https://docs.snowflake.com/en/user-guide/security-mfa-rollout ; USERS view https://docs.snowflake.com/en/sql-reference/account-usage/users ; LOGIN_HISTORY view https://docs.snowflake.com/en/sql-reference/account-usage/login_history ; authentication policies https://docs.snowflake.com/en/sql-reference/sql/create-authentication-policy ; network policies https://docs.snowflake.com/en/user-guide/network-policies